ALTO Working Group Meeting -
Thursday, July 29, 2021 15:00-16:00 PST (22:00-23:00 UTC)
Chairs: Vijay Gurbani, Jan Seedorf, Qin Wu
(https://datatracker.ietf.org/meeting/111/session/alto)
Video
Session Intro & WG Status Chairs (8 minutes)
Slides: https://datatracker.ietf.org/meeting/111/materials/slides-111-alto-chairs-slides-111-alto-00-02

[Vijay] Any update on unified properties?
[Sabine] Did most of the updates. Discussion among authors about points of design. To be closed with AD next week.
[Richard] Performance metrics updated to -17.
[Martin] To clarify, the performance-metrics - the most important comments have been addressed. The cdni-request-routing not got to yet. Hope to get to it in August.

AD guidance on the next charter and future of the group Martin Duke (5 minutes)
5 minutes to talk about state of charter
Was a lot of proposed work items. Some were a bit “researchy”
Since then, many “deployments” shown to be proof of concept. That’s fine, but it doesn’t convince Martin that there are actual ALTO deployments. IETF not in the business of extending protocols that are not used.
A few points:

This charter is scaled back to close up loose ends on existing protocol. Should not take a long time. Give WG the chance to show that existing work can be deployed. If not the case, this might mean that we have reached the end for ALTO. We have not made it thru IESG, we will see how this charter progresses thru the IESG - it is possible it will be slimmed down still further.

There is a lot of interesting research going. A lot of the proposal were research. IETF not research, but IRTF is. So this work might fit into IRTF especially PANRG. Hence Spencer and Jen on the agenda today. So please consider taking your work there to make progress (while ALTO catches up with you).

Any questions or comments, now is a good time.

[Richard] Great comment. I know of some reasonable deployments of ALTO in real life. They need to be documented.

[Martin] Glad to hear. Would be valuable to have actual deployers turn up to motivate use cases. We need to take time to do that. The work items originally proposed for the agenda struck me as a research agenda, and this is not the right venue.

ALTO and PANRG - PANRG chairs (15 minutes)
Slides: https://datatracker.ietf.org/meeting/111/materials/slides-111-alto-ietf111-alto-panrg-00

[Adrian] Looking at Slide 7 (How PANRG Could Help) I see the extremes of discussing experiences of ALTO and how that feeds into PANRG, and discussion of future use cases, but what is your view on developing protocol extensions or tweaks in PANRG?
[Jen] We are not doing protocol maintenance, but…
[Adrian] The difference I heard you say is between a Standards Track and Experimental document
[Spencer] Thats right. PANRG can publish Experimental RFCs, but if you need a Standards-track document, you need to be in the IETF.
[Vijay]
[Jen] Because we do research we try not to develop solutions, thats really the IETFs job.
[Richard] Could we investigate what ALTO tools did right, and what went wrong?
[Martin] Answering Adrian’s question ??? Once there is research consensus that an approach is viable, it can be taken to the WG and worked on. In general WGs don’t work on complex problems.
[Luis] ALTO is how, and it looks at the existing mechanisms available while PANRG explores new capabilities that could then be exposed by ALTO. Considering new scope, I see ALTO helping with the automation mechanisms.

Discussion of Working Group Charter (30 min)
Chairs to introduce 10 minutes
New Charter Open Discussion 20 minutes
[Martin] I have a little early feedback from the IESG. A lot of stuff there is “standard” and not normally mentioned in the charter. Typically the charter focuses on the document output (2nd and 3rd items here). There is special emphasis on deployment experience - would be worth IETF resources for this.
The new use cases requires additional thought, if we say here is another use case out in the Internet, might not be the most ideal way to find problems.
It would be really nice, if there are designs behind these use case, to have IETF put in resources. CDNI is interesting case. ALTO originally peer-to-peer: didn’t take off. Then tried CDN. It would have been better to have CDN providers turn up and say, yes this is what we want to fix.
[Vijay] To clarify: ALTO did succeed for peer-to-peer, but peer-to-peer itself did not succeed.
[Martin] Thats true, I dont mean to cast aspersions.
[Vijay] Peer-ro-peer use case was interesting when we started, but the use case went away.
[Richard] Question for Martin, these use cases might be backed up by deployments. There are papers that back up some of these, and I have spoken to authors. When I was talking to them they can clarify what is a real use.
[Martin] I am not advocating that they need to fly to Madrid [IETF112] to justify their use case, but having some way of showing that these use cases are beyond research.
[Luis] Highlight deployment at scale. Now new capabilities are being deployed with all the pieces.
There is a collection of IETF tools, programmatic APIs, YANG, PCE, et al. that provide automation. How these can be use for CDNs and ALTO can be used for automation.
A few words about CDN. CDN owned by operators (cf Telefonica proof of concept). Distinguish with interconneciton of CDNs. Note that discussions in CDNI this week (IETF111) consider ALTO for capacity, footprint, etc capabilities in situations driven by the Streaming Video Alliance.
Use cases such as compute etc. will be made possible by ALTO and not available with today’s tools.
[Martin] I will let other people take this discussion to the list.
[Cheng Zhou] Make case for network topology Based on our analysis CDN is built on top of an underlying network. We use BGP-LS and wonder if NETCONF can be used to collect data, with ALTO to compute CDN path.
[Peng Liu] Proposed draft for compute-aware. Has attracted attention. Maybe ALTO can do something with this. May help select service nodes and inform end points.
[Wei Wang] LMAP measurement exposure use case. As network operators we want to know better what happens to our network. LMPA provides good tool to collect measurement results. But results in file are not convenient. Our consideration is to use ALTO for aggregated network perfromance results, and the data can be aggregated as needed to get the desired network performance results.
[Yuexia Fu] I want to explain computing-aware use case. Hope to extend ALTO to support query of compute and network resources. Apply to static and dynamic information.

Wrap up: (2 minutes)